refactor(query): preserve rate/sum_over_time distinction in analyzer (kill PromQL string re-parse from PR #292)

[zzylol]

Summary

PR #292 fixed a real bug — the engine couldn't tell rate(metric[r]) from sum_over_time(metric[r]) because the analyzer collapses both to Capability::ExactAgg(Sum). The fix was a query_contains_rate_call walker in the engine that re-parsed the raw PromQL string at dispatch time to disambiguate.

That worked but was a lossy-lowering smell: the analyzer is supposed to be the single source of truth for query intent, and its lowering should preserve enough info that the engine doesn't need to re-parse.

This PR carries the distinction through the analyzer's typed output:

• Add OuterFn { Plain, Rate } to sketch_algebra::capability. Rate means the original PromQL had rate(...) / irate(...) anywhere in its tree (possibly nested inside an outer sum by (...) (...)).
• Add outer_fn: OuterFn to ASAPTierCandidate. The analyzer's existing trace_from_promql walker populates it in the same pass that captures function name / args / range.
• Engine's reducer dispatch in both execute() and execute_range_promql_modern() reads candidate.outer_fn instead of calling a query_contains_rate_call(query) helper.
• Delete query_contains_rate_call (~42 lines including doc).

try_topk_over_rate_fallback remains — it's a different concern (shape-extraction for topk-over-rate where the analyzer emits FrequencyTopk candidates the ExactAgg(Sum) sids don't satisfy). That's a structural analyzer-side change deferred to a follow-up; this PR's scope is the lossy-lowering smell only.

Test plan

• cargo build -p data_plane: clean
• cargo test -p data_plane --lib: 745 passed (was 743 post-feat(query): compose rate-over-Sum into ExactAgg(Sum) candidate (multinode topk dispatch) #292; +2 new regression tests). 0 failures.
• cargo test -p control_plane --lib: 745 passed (was 738; +7 new analyzer tests). 0 failures.
• Smoke test: all 5 spec queries return non-empty data_source: asap_query results:
  • rate(http_requests_total[5m]) → 4 zones, per-second rates (~20-22)
  • sum by (zone) (rate(http_requests_total[5m])) → 4 zones, per-second rates
  • topk(5, sum by (zone) (rate(http_requests_total[5m]))) → 4 zones (top 5, K ≥ zone count)
  • sum_over_time(http_requests_total[5m]) → plain reducer, raw window sum (25988) — load-bearing regression case
  • quantile_over_time(0.99, http_requests_total_latency_ms[5m]) → 4 zones p99
• rg "query_contains_rate_call" data_plane/src/ → only doc-comment references documenting the retirement; no function call sites.

Regression test additions

Analyzer-side (control_plane::asap_tier_analysis::tests):

• rate_candidate_carries_outer_fn_rate
• irate_candidate_carries_outer_fn_rate
• sum_over_time_candidate_carries_outer_fn_plain — the critical "shares capability with rate, MUST report Plain" case
• sum_by_candidate_carries_outer_fn_plain
• bare_selector_candidate_carries_outer_fn_plain
• sum_by_over_rate_candidate_carries_outer_fn_rate — composed shape: outer fn name is sum but inner rate triggers OuterFn::Rate
• rate_and_sum_over_time_share_capability_but_differ_on_outer_fn — pins the asymmetry the engine's dispatch reads off

Engine-side (data_plane::query_engines::asap_query_engine):

• execute_sum_over_time_dispatches_to_plain_exact_agg_reducer — pins the per-window reducer's output (sum of raw window values, NOT divided by the range — if the dispatch ever regresses to "all ExactAgg(Sum) + range > 0 → rate path", the asserted value changes by a factor of range_seconds).
• analyzer_candidate_outer_fn_distinguishes_rate_from_sum_over_time — documents the typed contract the engine reads off.

Smoke-test outputs

=== Q1: rate(http_requests_total[5m]) ===
{"result":[{"zone":"z0","val":"22.53"},{"zone":"z1","val":"22.52"},{"zone":"z2","val":"20.79"},{"zone":"z3","val":"20.79"}],"infos":["data_source: asap_query"]}

=== Q2: sum by (zone) (rate(http_requests_total[5m])) ===
{"result":[{"zone":"z0","val":"22.53"},{"zone":"z1","val":"22.52"},{"zone":"z2","val":"20.79"},{"zone":"z3","val":"20.79"}],"infos":["data_source: asap_query"]}

=== Q3: topk(5, sum by (zone) (rate(http_requests_total[5m]))) ===
{"result":[{"zone":"z0","val":"22.53"},{"zone":"z1","val":"22.52"},{"zone":"z3","val":"20.79"},{"zone":"z2","val":"20.79"}],"infos":["data_source: asap_query"]}

=== Q4: sum_over_time(http_requests_total[5m]) ===
{"result":[{"metric":{},"val":"25988"}],"infos":["data_source: asap_query"]}

=== Q5: quantile_over_time(0.99, http_requests_total_latency_ms[5m]) ===
{"result":[{"zone":"z3","val":"59.74"},{"zone":"z1","val":"83.94"},{"zone":"z2","val":"70.12"},{"zone":"z0","val":"90.94"}],"infos":["data_source: asap_query"]}

🤖 Generated with Claude Code